New book · 2026
Teaching in the AI Era: Practical AI Strategies for Teachers
A chatbot can now write your students’ homework in seconds. It cannot do their learning for them. What should a teacher do?
Open to Work
Applied AI & Solutions Engineer
Agentic Systems · LLM Integration · Workflow Automation · Production Deployment
I build and deploy agentic AI systems that automate real workflows: from multi-agent orchestration and MCP integrations to full-stack applications running in production.
Barcelona, Spain · Open to Applied AI, AI Solutions, Automation and Forward Deployed Engineering roles
Founder of LoopCodeLab · PhD Researcher at UAB Barcelona
What I do
I work end to end: I scope the real problem, design a practical AI solution, build the full-stack application, integrate the models and services, and deploy and operate it on my own Linux infrastructure.
Understand the problem. Turn operational and user needs into a clear, buildable AI solution.
Design AI solutions. Agentic workflows, multi-model routing, LLM-as-a-judge evaluation, tool calling and structured outputs.
Build full-stack. Complete applications across Python, Node.js, React and FastAPI.
Integrate everything. LLM APIs, REST APIs, MCP servers, WhatsApp, email and calendar, and automation tools like n8n.
Deploy and operate. Ship to Linux with Docker, Nginx, PM2 and systemd, with TLS, monitoring and failure handling.
Communicate clearly. Explain technical systems to business, research and education stakeholders so they get adopted.
About
I am an Applied AI and Solutions Engineer based in Barcelona. I design, integrate, evaluate and deploy production AI systems: agentic workflows, multi-LLM evaluation, real-time voice AI, MCP tools, AI media pipelines, and the full-stack applications and self-hosted infrastructure that run them.
My background combines engineering execution with doctoral research and more than ten years across education and technology. That mix lets me translate cleanly between business needs, user workflows and technical architecture, and to ship systems that non-technical teams can actually adopt.
Independent · Since 2024
Design, build and deploy production AI systems and the infrastructure they run on
Self-hosted multi-agent platform that plans, builds, reviews and ships software from a single prompt
AI education platform generating curriculum-aligned lesson plans and quizzes, built for Urdu and offline use
Universitat Autònoma de Barcelona
Pedagogical digital competence and AI adoption in higher education. Supervisor: Dr. Cristina Mercader
Author of 15 published Urdu columns on life, learning and society
Featured Case Studies
Five production systems that show the full arc: problem, architecture, integration, deployment and operation. Most run right now on infrastructure I manage myself.
A self-hosted platform where planner, worker and reviewer agents turn a plain-language requirement into planned stories, parallel implementation, code review, Git integration and deployable software. Agents build in parallel on isolated git worktrees, a reviewer merges each change, and one idea ships to web, mobile and the app stores. Bring your own AI keys and own every line.
Parallel agents over tmux and Git, with delivery to the App Store, Google Play and the Microsoft Store
A remote MCP server that lets Claude watch, task and unblock a team of xAI Grok Bots on the cloud computer they share. Grok Bots have no API, so Claude reads each agent's activity and transcript in place, delegates through a per-agent inbox and outbox, clicks on the agents' desktop when one is stuck, and generates images and video with Grok Imagine. Designed by Claude, written by the Grok CLI and reviewed by the Codex CLI.
Usable from claude.ai on the web or a phone, behind Google sign-in with a single allowed account
An MCP server that turns a grocery's product catalogue into conversational tools. Staff scan a barcode and say what they want in plain language, and the assistant searches the catalogue, looks the EAN up in Open Food Facts, drafts the entry in the house naming style, applies the correct Spanish IVA rate and queues the approved product for the shop's existing TPV. The till stays the system of record. TillBridge only feeds it.
Augments a legacy point-of-sale terminal rather than replacing it, with every write confirmed by a person first
A real-time AI assistant that answers WhatsApp voice calls and holds a natural conversation. Paco checks email, calendar, weather and the web mid-call through live tool calls, and places outbound briefing calls through a simple HTTP API. Built from an empty folder to production in a single day.
Deployed on my own Ubuntu server behind Nginx, with bidirectional audio streaming and a custom operations dashboard
A free, open-source MCP server that gives AI assistants local text-to-speech and voice cloning. No API keys, no cloud, running entirely on your machine. Listed on mcpservers.org and consumable by any MCP client.
Compatible with Claude Desktop, Claude Code, n8n and any MCP client
More Projects
A wider set of AI products, developer tools and integrations. Where a demo is public, the card links straight to the live system.
AI education platform that generates curriculum-aligned lesson plans, quizzes and personalised learning content for teachers and schools. Designed for Pakistani classrooms: Urdu support, print-friendly output and affordable delivery.
Computer-Assisted Iterative Thematic Analysis: a three-stage qualitative-analysis workflow with structured coding, Excel exports and transparent researcher control. Published software with a DOI on Zenodo.
The self-hosted engine behind LoopCodeLab. A browser and phone workspace where AI agents plan, build, review and ship whole projects in parallel, over xterm.js and a tmux PTY bridge. Live behind this link.
Transcript-first AI video editor: upload an interview, strike a line, and those seconds leave the cut. A hosted, mobile-first studio powered by OpenAI transcription and Remotion rendering across a seven-service Docker stack. Live behind this link.
LLM evaluation studio: switch between Claude, GPT, Gemini, Groq, Qwen and DeepSeek, or compare two side by side with an automatic judge. Provider switching, error handling, retries and usage monitoring.
Reference manager for my PhD thesis: 590+ sources with client-side semantic search over embedded vectors and exports for citation tools. Live demo.
Published npm package. Turns free text into a short, DNS-safe slug, safely fitting a clean project name into a single 63 character hostname label (RFC 1035). Zero dependencies.
Flutter build and distribution pipeline on a headless server: hot-reloading web previews behind TLS and signed Android release APKs, plus a shipped cross-platform game.
Custom n8n node connecting WhatsApp to automation workflows, running in a self-hosted Docker n8n stack alongside AI pipelines.
Self-hosted, customised deployment of the OpenClaw open-source assistant, running as a daily-driver AI system on my own infrastructure.
A remote MCP server that lets Claude operate the cloud computer shared by a team of Grok agents: shell and files, image and video generation, each agent's activity and a task inbox, and the desktop. It reaches the machine only through a reverse SSH tunnel, behind Google sign-in, with a pinned host key, a per-command identity check and an audit log of every call.
A Proxmox virtualisation host with MCP connectors that give Claude a Debian desktop, a Windows machine and disposable sandbox VMs cloned from a golden image with Claude Code, Codex and Gemini preinstalled. Sandboxes sit on a walled network behind an allow-list proxy and are destroyed automatically when their lifetime ends.
AI application that generates, manages and grades mathematics problems with personalised student feedback.
This site and its sister services are self-hosted on an Ubuntu server I administer: Nginx terminates TLS for more than a dozen subdomains, Node services run under PM2 and systemd, and n8n runs in Docker. A second, Proxmox-based host runs isolated virtual machines for AI agent sandboxes. The portfolio is the proof.
Technical Capabilities
Next milestones: structured LLM-evaluation harnesses, retrieval over embedded corpora (already used client-side in RefLib), and richer observability across my agent workflows.
Experience
More than ten years across education and technology, including recent experience building and deploying AI and EdTech products. My engineering and my education work are shown separately below.
AI Engineering & Product
LoopCodeLab & AI Product Portfolio · Barcelona
PakEdX · Remote
Education & Research
Universitat Autònoma de Barcelona · Barcelona
District Education Authority, Gujranwala
Public Education Sector
EdTech Specialization
EdTech teams need engineers who understand classrooms as well as code. I bring both: education-sector experience, doctoral research and shipped AI products that teachers actually use.
Education-sector experience. Teaching, instructional design and district-level roles across public education.
Doctoral research. Pedagogical digital competence and AI adoption in higher education at UAB.
Curriculum & assessment. Real understanding of curriculum alignment, assessment and teacher training.
Evidence-based design. Digital-competence and learning-analytics research feeding practical product decisions.
AI product development. PakEdX lesson planning and assessment, plus AI_Teacher for maths practice and feedback.
Full-stack delivery. I build and deploy, so an EdTech idea reaches real teachers and students.
Research
Thesis: Exploring university teachers' pedagogical digital competence and its relationship with teaching practices. Supervisor: Dr. Cristina Mercader. My research supports evidence-based product design, user research and responsible AI adoption.
A mixed-methods study of university teachers, combining statistical and qualitative analysis to identify adoption barriers and effective practices for AI-enabled learning.
New book · 2026
A chatbot can now write your students’ homework in seconds. It cannot do their learning for them. What should a teacher do?
Ilyas, M. T. (2026). AI-supported lesson planning for teacher education: A capacity-building pilot at the University of Gujrat. Facilitator manual (2nd ed.) [Lesson]. Zenodo. https://doi.org/10.5281/zenodo.24397014
Ilyas, M. T. (2026). AI-supported lesson planning for teacher education: A capacity-building pilot at the University of Gujrat. Participant workbook (2nd ed.) [Lesson]. Zenodo. https://doi.org/10.5281/zenodo.24397088
Ilyas, M. T., Mercader, C., & Bosco, A. (2026). From technical literacy to AI-integrated pedagogy: A scoping review of digital competence frameworks for higher education (2000–2026). The Turkish Online Journal of Educational Technology, 25(4), 56–78.
Ilyas, M. T., Mercader, C., & Bosco, A. (2026). Infrastructure-first digital competence in resource-constrained higher education. Jurnal Pendidikan dan Profesi Keguruan (Progresif), 6(1), 77–91.
Ilyas, M. T., Mercader, C., & Bosco, A. (2026). WhatsApp as shadow learning management system: How university teachers in Pakistan build parallel pedagogical infrastructure under institutional prohibition. ACADEMIA International Journal for Social Sciences, 5(3b), 343–354.
Ilyas, M. T., & Zaman, S. (2023). An empirical enquiry into the teachers' philosophical beliefs and their actual classroom practices: A case of Pakistan. Journal of Positive School Psychology, 7(4), 1843–1853.
Ilyas, M. T. (2026). Exploring university teachers' pedagogical digital competence and its relationship with teaching practices [Doctoral thesis, Universitat Autònoma de Barcelona]. TDX.
Ilyas, M. T., & Mercader Juan, C. (2026). What drives successful technology integration? An examination of teacher digital competence through DigCompEdu in Pakistan. In M. J. Bolarín Martínez & A. Bernárdez Gómez (Eds.), Libro de actas del XIV Congreso Iberoamericano de Docencia Universitaria (pp. 320–321). Universidad de Santiago de Compostela.
Ilyas, M. T. (2025). CAITA: Computer-Assisted Iterative Thematic Analysis (Version 1) [Computer software]. Zenodo.
AI-Supported Lesson Planning for Teacher Education: A Capacity-Building Pilot at the University of Gujrat
Credentials
Rust Foundation
Credibility
Published MCP server
Listed on mcpservers.org, MIT licensed
npm package
@tayyabcheema777/smart-name, zero dependencies
Software DOI
CAITA on Zenodo, 10.5281/zenodo.17474907
Peer-reviewed publication
Scoping review accepted at OJET, July 2026 issue
Research grant
UAB Fons de Solidaritat, EUR 1,500 (2026), project lead
15 published columns
Urdu columns at Daily Urdu Columns
Writing
Alongside engineering, I write a regular Urdu column at Daily Urdu Columns, with 15 published pieces on life, learning and society.
Writing keeps my core professional edge sharp: explaining complex ideas simply. That is exactly what turns an AI concept into something a non-technical team will understand, trust and adopt.
Read the columns →I also keep an engineering blog on MCP connectors, multi-agent systems and the infrastructure under them, written from systems that run in production.
Read the blog →Contact
Looking for an engineer who can turn an AI concept into a deployed working system? Let's talk. Based in Barcelona, Spain, and open to Applied AI, AI Solutions, Automation and Forward Deployed Engineering roles.